Guide · The back office, on autopilot

The Small Business Guide to the Model Context Protocol

A plain-English guide to the standard that could end back-office drudgery for good.

If you run a small business, a community venue, or a busy professional practice, you are probably the real integration layer of your organisation. Here is what MCP is, how it works, and how to put your operations on autopilot.

Every Tuesday night or Sunday afternoon, you sit in the middle of a digital switchboard. The booking calendar knows who reserved the rooms. The invoicing package knows who owes what. The inbox is where you send the gentle chasers. None of these systems talk to each other automatically, so you copy the name from one screen, paste it into the next, compute the VAT, generate the checkout link, and draft the email.

It is not difficult work, but it is relentless. It eats your evenings and creates a constant background hum of administrative stress.

For years, software companies promised to solve this with "native integrations." They told you that if you just bought their specific suite, or paid for a complex no-code automation platform, the data would flow. In practice, you ended up paying extra subscription fees for brittle connections that broke the moment a provider changed their API.

Now, a new technology standard is quietly changing how business software connects. It has the potential to eliminate the drudge work of back-office administration forever.

What actually is the Model Context Protocol?

To understand MCP, think about the history of home appliances. Before the standardised electrical wall outlet, if you bought a new lamp or toaster, you had to hire an electrician to hard-wire it directly into your home's electrical system. Every appliance needed a custom connection.

The wall outlet changed everything. It established a standard socket. Appliance makers built a plug, home builders installed the socket, and the user could plug anything into the wall without knowing a thing about electrical engineering.

MCP is the universal wall outlet for artificial intelligence.

Historically, if you wanted your AI assistant, like Claude or ChatGPT, to read your booking diary or write an invoice, developers had to build a custom, one-off integration between that specific model and that specific software API. It was expensive, slow, and insecure.

With MCP, every software tool publishes a standard "plug." The AI assistant plugs into that socket and instantly understands how to read and write data in that system.

The AI as the operator in the middle

The magic of MCP is that the AI assistant does not act as a passive data pipe. It acts as an active, logical coordinator. Connect it to several MCP-compatible tools, your booking calendar, your accounting package, your email, and the AI becomes the switchboard operator. It listens to your instructions in plain English, works out which tools it needs, and coordinates the work between them.

You type

Check Doris for last month's unpaid bookings, draft a polite reminder to each client in my usual tone, and leave them in my drafts folder.

The AI doesn't run a rigid script. It reasons through the task:

  1. 1Queries the Doris MCP server to list bookings and identify which ones are unpaid.
  2. 2Processes the list of names and email addresses.
  3. 3Drafts the personalised emails in your tone of voice.
  4. 4Calls your email MCP server (Gmail or Outlook) to save the drafts.

You did not map a database field or test an API connection. You described the job in ordinary English, and the AI handled the wiring.

One connector is a party trick. Five is a different animal.

Having your AI read your calendar is a minor convenience. The real transformation happens when you connect several tools to the same switchboard, and you don't configure a different integration for each job. You explain what you want, and the same operator executes it.

DorisQuickBooksStripe

The invoicing pipeline

Match room usage to client invoices, generate Stripe checkout links, and update invoice memos automatically.

DorisMailchimp

The marketing engine

See who booked last term, spot which customer groups are growing, and draft a tailored newsletter campaign.

DorisGoogle Sheets

The budget advisor

Extract your real utilisation numbers, build a spreadsheet model, and forecast next year's revenue under different pricing.

Why "pencil, not pen" is the golden rule

It is natural to feel nervous about letting an AI near your accounts or email. Models make mistakes, and risks like "prompt injection," where a malicious email hides instructions to trick the AI, are real. So well-designed MCP integrations follow a strict rule: work in pencil, never in pen.

The AI does the heavy lifting of reading, analysing and drafting, but it is not allowed to press the button that counts.

The reminder email is written, but it waits in your drafts folder until you click send.Draft
The invoice is prepared, but it sits as a draft in QuickBooks until you approve it.Draft
The payment match is suggested in Doris, but it sits in a review strip until you confirm.Review

The AI is physically incapable of moving money, because the payment tools do not expose a "send money" lever to the assistant. The AI prepares the paperwork; the human keeps the judgment and presses the final button. This simple boundary turns a potential AI mistake from a business crisis into a caught typo.

Getting started, three simple components

Connecting your business to an MCP-enabled assistant takes three pieces. Once configured, the AI runs the handshake, discovers the tools, and is ready for your instructions.

01

The brain, the client

An MCP-compatible AI assistant. The most common options today are Claude and ChatGPT, both of which can connect to MCP servers out of the box.

02

The plugs, the servers

The software you use must publish its own MCP endpoints. A growing number already do:

Dorisapi.bookwithdoris.com/mcp
QuickBooks OnlineMCP server available
StripeMCP server available
ZapierBridges thousands of apps
03

The credentials

You authorise the connection through a standard, secure OAuth consent screen, similar to logging in with your Google or Microsoft account.

The economics: a multiplier, not a replacement

Some AI companies promise their tools will let you "replace your staff." That is a misleading and unhelpful claim. A good administrator does forty things that require human relationship-building, empathy, and localised judgment.

What an MCP-connected assistant actually does is take the rules-based drudgery off their plate. By automating the copy-pasting, the double-checking, and the initial drafting, you free up your team, or yourself, to focus on the relationships that build your business.

Doris keeps the diary honest and publishes what it knows. The operator in the middle carries it the last few steps, in pencil. And you get your Tuesday evenings back.

See what a switchboard that runs itself feels like

Doris is, as far as we can tell, the first room-booking platform to ship with a native, secure MCP server. Up to 28 days free, no card required.

Start your free trial